pollutionspots:anovelmethodfor air pollution monitoring · bogota, colombia abstract ... means of...

Pollution spots: a novel method forair pollution monitoring

L. Rodrı́guez & D. MendezDepartment of Electronic Engineering, Pontificia Universidad Javeriana,Bogota, Colombia

Abstract

The Participatory Sensing paradigm offers the advantages of leveraging theexisting communication networks and mobile phones (used as sensing devices),thus reducing the costs of deploying a sensing network. Despite the processingcapabilities and the already embedded sensors in smart phones, integratingpollution sensors into cellphones is still an ambitious project. Pollution-Spotsproposes a solution by means of using an infrastructure of fixed low-costsensing devices, and reporting the measurements using Participatory Sensing. Thesystem’s name comes from the static sensing units which function as spots themobile user can hook-up to in order to obtain the environmental data, generatinga model that easily escalates, since it does not require users carrying the sensingdevices. If the system’s administrator requires to enlarge the region of interest,only the installation of an additional Pollution-Spot is required. The device startsmeasuring and the pedestrian forwards the information, completing the cycle withno extra cost of data transport and/or human resources. Preliminary experimentshave been conducted using the platform located in Pontificia UniversidadJaveriana’s campus. The maps obtained based on the gathered data provide abetter spatial resolution of the variables of interest when compared to traditionalmonitoring stations, proving the effectiveness of the system. These results showthe benefits of applying this model in areas with scarce monitoring stations and/orbudget limitations. Also, including user participation makes it possible to designapplications that improve communities’ quality of life, since the maps obtainedallow the identification of the areas where the concentration of pollutants might becritical.Keywords: air pollution, participatory sensing, environmental data visualization.

Air Pollution XXIII 129

www.witpress.com, ISSN 1743-3541 (on-line) WIT Transactions on Ecology and The Environment, Vol 198, © 2015 WIT Press

doi:10.2495/AIR150111

1 Introduction

Participatory Sensing (PS) is a sensing paradigm that takes advantages of theadvances and ubiquity of mobile technology, as well as the cellular data networkinfrastructure already deployed, to create large scale sensing schemes [1]. ThePS based schemes cover different data types, mainly information about theenvironment, weather and traffic congestion, by means of groups of cellular users,that contribute with the measuring and the retrieving of information. The collecteddata are centralized and are generally used to increase collective knowledge. In thecase of environmental data, and more specifically the data concerning air quality,the sensing variables (such as temperature, atmospheric particulate matter (PM),and the concentration of different gases) can be used to visualize their behaviorover the time-space region of interest, thus helping communities to increase theawareness about pollution, and consequently improving their quality of life.

One concern related to the usage of PS schemes to measure air quality, is thefact that despite the processing capabilities and the embedded sensors in smartphones, these smart phones still do not include the required sensors to acquire thevariables of interest, pollutants in this case. Since attaching these sensors to themobile devices is a burdensome project, it is necessary to come up with a schemethat overcomes this obstacle and, at the same time, exploits the benefits of PS,i.e., massive coverage of a large number of phenomena implied by the mobility ofphone carriers, the inclusion of people in the sensing loop, and no costs in terms ofdata transport. Considering this context, in this paper we propose Pollution-Spots,a novel scheme for measuring environmental data, that proposes a solution, bymeans of an infrastructure of static low-cost air quality sensing devices, insertedinto a PS-based scheme, combining user participation to transmit and centralizethe data, so as to avoid the extra cost of data transport and/or human resources.The result is a cost-effective air quality sensing system that is easy to deploy andescalate.

The name of the system comes from the sensing stations, which are static andtherefore can be seen as spots that the user can hook up to in order to obtain thesensing data. In this model, it is easy to enlarge the region of interest, since it onlyrequires the installation of a new Pollution-Spot to acquire the environmental dataof that extended region. The data are collected by the pedestrian, who forwards theinformation using the already deployed wireless data networks (cellular carriers,Wi-Fi, etc.), centralizing this information so that it can be processed and madeaccessible through a web server.

The rest of this paper is organized as follows: we discuss the related work inSection 2. In Section 3 we detail the proposed scheme, depicting the system’sarchitecture and main functions of the nodes within it. Section 4 describes thechallenges associated with the deployment of a PS based scheme and the solutionsproposed by the Pollution-Spots model. Section 5 refers to the preliminary resultsobtained in a case study located in the campus of the Pontificia UniversidadJaveriana. Finally, we conclude the paper in Section 6, where we also brieflydiscuss the guidelines to continue our work.

130 Air Pollution XXIII


2 Related work

There are different examples in literature of applications that sense environmentaldata exploiting the characteristics of PS schemes. The work depicted in theseexamples has a common denominator: the concept of the implied mobility ofthe sensors, such that they are at all times attached to the mobile devices. Thisnotion has the disadvantage of requiring trust of the users with the sensing devices,reducing the possibilities of obtaining more granulated data, since there is a limitednumber of trusted users, and an even more limited number of sensing devices.This approach also has the disadvantages of requiring multiple sensing tasks (atdifferent time-location points) from the user, a factor that also reduces the numberof users willing to collaborate with the system, no matter the involved incentives.Such is the case with Haze Watch [2], which relies on air pollution sensors attachedto motor vehicles to perform the readings. The sensing data are sent to the databaseusing iPhones, which also have to be attached to the motor vehicles.

A similar system, Common Sense [3], has developed hand-held monitors withair quality sensors as a prototype, in order to resolve the absence of the sensorsin consumer devices. The monitors can communicate with the database throughBluetooth, 802.15.4, or GPRS radios, depending on either user interest or datafidelity. This project also addresses the challenge of making the monitors ascomfortable as possible to collaborators in order to facilitate the measurementtask. Another example is P-Sense [1], a system where the environmental data arecollected by a set of individual external sensors integrated in a board. The sensingdevices can be carried along with the cellphone or function as standalone devicesdue to their independent power supply. The proposed architecture of P-Sense isused as a reference point for Pollution-Spots.

Concerning the transport of environmental data to the final user, these deliverytechniques commonly use a web server that displays the variables of interestthrough maps, charts, tables, etc. The work in [4] proposes not only thevisualization of raw sensor data, but also contour maps. In this project the sensorsare not distributed in a regular fashion, raising the need of clustering the sensorswhen displaying them in maps, due to the differences in sensor resolution betweenareas; this is not a concern with our project, since we utilize spatial interpolationtechniques that do not require regularly located sensors.

The project in [5] proposes the use of marker maps to display informationin specific time-space locations and heat maps with the available measures in aregion. They also display tables with information as user request them, by means ofexploiting cloud storage services such as Google Fusion Tables for processing anddelivering the data. Another interesting idea can be found in [6], where the authorsuse hierarchical and time-series data to display and compare the correspondingdata of each layer. Applying this method with air pollution data helps to get abetter perception of the behavior of the variables of interest and discriminate thesource of the pollution.

The project Minutely [7], though working with forecast data (instead of rawdata as intended in this project), provides a good example of how to integrate



people in the sensing loop. Minutely combines the weather radar from the Bureauof Meteorology in the United States and Australian areas, with personal reportsfrom the participants, to deliver real time weather forecasts.

3 Proposed scheme

Figure 1 depicts the proposed scheme for Pollution-Spots system’s architecture.The model is composed by five layers (from lower to higher level): the DataCollection Layer, the Data Analysis Layer, the Data Network Layer, the DataStorage and Feedback Layer, and the Data Storage, Estimation and VisualizationLayer.

Figure 1: Pollution-Spots system’s architecture.

3.1 Data Collection Layer

The first layer is the Data Collection Layer, which is composed of a network ofstatic sensors. Each sensing device has a set of environmental sensors integrated ina board (based on an AVR Atmel processor), so they function as a single unit. Thesensing features of the units include measurement of ozone (O3), relative humidity(RH), temperature (T ), particulate matter (PM ), and carbon monoxide (CO).Table 1 lists the embedded sensors in the sensing units.

The devices must work as standalone units. For this reason they include anindependent power supply based on an energy harvesting module and a solarpanel, making the solution autonomous and sustainable. The set of sensors collectsmeasurements at regular time intervals. The collected data, with the corresponding



time stamps, are packed in a single frame. The frames are stored in the sensingunit until the mobile user requests them through the mobile sensing application inher cellphone. When requested by the application, the frames are transmitted viaBluetooth to the mobile cellphone and erased from the local storage in the sensingunit.

Table 1: Sensors embedded in the sensing units.

Manufacturer – Sensor Variable

Membrapor – CO/SF-200 CO

Membrapor – O3/S-5 O3

Sensirion – SHT15 Temperature and Humidity

Sharp – GP2Y1010AU0 Particulate Matter

3.2 Data Analysis Layer

Mobile phones compose the second layer of the system. In hands of theparticipants, they act as a point of gathering, analysis, and forwarding of theacquired data. Since the mobile devices transmit the data using the existingwireless data networks (mobile data carriers, Wi-Fi, etc.), the scheme has noextra costs in terms of data transport. The mobile application is developed underAndroid: the popularity of smart phones based on this operating system increasesthe possibilities of attracting more users to collaborate with the project.

The mobile application communicates with the Data Storage and FeedbackLayer, and the Data Storage, Estimation and Visualization Layer. When the userreceives a request for measurements, the application connects to the requiredPollution-Spot and receives the data from the sensing device (located in thecoordinates specified by the request). The application attaches a GPS-basedlocation stamp to the environmental data and forwards it to the Data Storage andFeedback Layer.

3.3 Data Network Layer

The third layer of the Pollution-Spots system is the wireless network infrastructurebased on the IP technology (mobile data carriers, Wi-Fi, etc.) already deployed. Bymeans of this layer, the users forward the sensed data to the higher levels of thearchitecture. The user is allowed to decide whether to use the data cellular networkor Wi-Fi, and also when to send the data to the centralized server. Since this modelof data transportation is subject to the perils inherent to a public data network, the



development of a privacy preserving mechanism to protect the critical user data isrequired.

3.4 Data Storage and Feedback Layer

The Proxy Server composes the Data Storage and Feedback Layer, and actsas an intermediary between the Data Storage, Estimation and VisualizationLayer and the mobile phones in the Data Analysis Layer. This mediation isnecessary to guarantee an extra layer of protection for the participants’ privateinformation [8, 9], which will be discussed later on. The server also managesa database (based on MySQL) with the data from the participants. The userdata quantify the participation of each participant, which can be used to provideincentive mechanisms to maintain the required amount of users in the system. TheProxy Server forwards only the environmental data with the time-location stampto the Pollution Data Server, disengaging user data from environmental data.

3.5 Data Storage, Estimation and Visualization Layer

The Pollution Data Server, located in the Data Storage, Estimation andVisualization Layer, is the core of the Pollution-Spots network, and it implementsenvironmental data storage and management. The stored data are used to createthe maps of the variables of interest. The management of the pollution dataincludes algorithms for data interpolation, privacy protection, incentives, as wellas verification of the measured data quality. These algorithms will be discussed inthe following section.

4 Challenges of PS based systems

Since the proposed scheme is based on a PS system, it is necessary to addressthe inherent challenges of this type of sensing paradigm. First, since the systemdepends on a public wireless network, user location data could be compromised.This problem rises the question: How to implement techniques to protect userprivacy? Another obstacle is the reduced budget for encouraging the crucial usercollaboration, since centralizing the collected data relays on it. Therefore, whichmechanisms are adequate for an air pollution monitoring PS system that has nofunds to compensate user participation?

In order to address these questions, Pollution-Spots implements a combinedalgorithm that protects user privacy and, at the same time, recognizes herparticipation. To achieve this, it is necessary to isolate environmental data providedby the user from her identity. This is the reason of the existence of the Data Storageand Feedback Layer. The proposed algorithm is based on the hybrid privacymechanism presented in [8] and commonly practiced gamification techniques tomotivate user collaboration [10]. To simplify the discussion about the algorithm,we first discuss the privacy mechanism and later the incentives mechanisms.



4.1 Privacy mechanisms

After the first collection of measurements, the Pollution Data Server in the DataStorage, Estimation and Visualization Layer interpolates the data points to obtaina data surface of each variable of interest. Each surface is processed to divide theregion of interest in areas according to the variability of the data (in terms of spatialcoordinates). The cell division algorithm is based on Voronoi decomposition, andit is performed on each environmental data collection as follows:

• At first, the region is divided in cells of uniform size.• After every iteration, the areas where the variability of the data is low are

merged to reduce the amount of data required to reconstruct the data surfacesince the exact location information is not very important.

• On the other hand, the algorithm splits the areas where the variable ofinterest has a higher variability, and for these regions the information ofthe exact location is critical.

Figure 2: Privacy Mechanism scheme.

With the division of the region of interest in areas of different sizes, the system isable to determine how the information about the data location will be delivered. Inlarge areas (low variability) the location of the data, which is associated to a user,is not so critical and therefore it can be anonymized using points of interest. Spatialcoordinates in smaller areas (larger variability), where it is important to know theexact location, must be encrypted to avoid compromising the user’s position. Thisdecision of whether to encrypt or anonymize the location data is the main ideaof the hybrid privacy scheme proposed in [8] (see Figure 2). Based on a previouscell’s division, the Pollution Data Server broadcasts requests to the mobile users,each request specifying the location and time window where the measurements areneeded, and the policy to protect the data.



4.2 Incentive mechanism

The database located in the Data Storage and Feedback Layer holds theinformation that quantifies the user participation in the system. This informationis essential when applying the incentives mechanism, because it is used to giverewards in form of points, badges and/or recognition. These techniques are a subsetof Game Mechanics, the name of the mechanisms used as building blocks forgamifying an application.

There are examples of PS applications that exploit gamification to reward theuser’s participation [11, 12]. Though the examples illustrate how to use gaming inthe context of PS, these works do not address the problem of relating the amount ofcooperation supplied from the user and, at the same time, maintain her privacy. Thework in [13] is an approach to this issue. The target of this work is applications withno budget constraints, and for this reason, it does not entirely solve the problemsaddressed here. Also, because of monetary compensations, the rewards are giveninstantly, so the system does not have memory of user collaboration, as needed toimplement gamification.

The proposed solution (depicted roughly in Figure 3) works as follows: Whena new measurement task is generated by the Pollution Data Server, the PollutionData Server assigns a label to it, known as request stamp. This label is calculatedthrough a Hash function, using a key k known to the mobile applications, but notto the Proxy Server. The label is associated to the amount of reward correspondingwith the mission, and it is sent only to the Proxy Server. The mobile applicationreceives only the Location-Data information of the request, but can generate therequest stamp since it knows the key k.

Figure 3: Combined Privacy and Incentive mechanisms in Pollution-Spots.

Once the user performs the measurement task, the application sends thepollution data to the Proxy Server. The application also sends the user ID andthe request stamp. If the Proxy Server finds a match between the request stamp



sent by a cellphone and the one sent from the Pollution Data Server, it assigns thecorresponding reward to the user. The user’s category is upgraded in the ProxyServer’s database with her new achievement.

The Proxy Server forwards only the environmental data (which do not containinformation about the user) to the Pollution Data Server. Since the environmentaldata (containing the location-time data) have been already treated with the privacypreserving mechanism, it is a hard task for a malicious entity to obtain informationabout the user while listening to messages between the Data Analysis and the DataStorage and Feedback Layers. Compromising the Pollution Data Server offers nobenefits in terms of user data, since the information stored in this server concernsonly to the variable of interest.

5 Results and analysis

To verify the performance of the sensors embedded in the modules that willfunction as Pollution Spots, tests where performed in an area of the city nearbya monitoring station of Bogota’s Environmental Agency (SDA – SecretariaDistrital de Ambiente). Due to the lack of specialized equipment and a controlledenvironment to calibrate the system, it has been a very difficult task to properlyvalidate the level of accuracy of the integrated sensors. However, our experimentsshow the feasibility of the modules, since the behavior of the measurements isconsistent to those provided by the SDA. The differences in the exact values areexplained partially by the quality of the sensors in the module and partially by thedifferences between the location of the measurements.

The data collected from the Pollution-Spots via the smart phones (which receivethe measuring tasks that indicate where and when to collect the data), have gapsin time and space, therefore an interpolation technique must be used to create amap to correctly visualize the variable of interest. The spatial data are interpolatedusing cubic splines, which was selected because cubic interpolation offers a goodtrade-off between the computational cost and smoothness of the resulting surface.As depicted in Figure 5, the system requires a minimum number of measurementsto perform a correct reconstruction of the data surface. The final interpolationgenerates a 38 × 38 data points matrix, which can be correctly generated using50 samples in the area of interest (refer to the right graph in Figure 5). However,with only 10 samples in the area of interest (center graph in Figure 5), the generalbehavior of the variable can be recognized.

The collected air quality information is available to the users by means of aweb server. The main functionality of the web server is the visualization of thedata in the region of interest as a partial retribution for the user participation. Theair quality data allow the user to get a better perspective of the behavior of thevariables in the area of interest.

As presented before, an initial version of Pollution Spots has been deployed inthe campus of the Pontificia Universidad Javeriana. Figure 4 shows how the finaluser can visualize the behavior of the variables of interest in the University campus.The webpage uses color gradient maps and a scale to facilitate the understanding



Figure 4: Original data surface for CO (left), reconstruction with 10 samples(center) and with 50 samples (right).

Figure 5: Carbon monoxide concentration (ppm) in the campus of the PontificiaUniversidad Javeriana.

of the maps. The figure clearly shows how the lowest concentration levels (red) areobtained in the green areas of the campus, meanwhile the highest levels (magenta)are near the roads and the parking building, which naturally exhibit a greaterautomobile traffic during the day.

The color data surface is updated with every new iteration (collected data) inthe Pollution Data Server, and its visualization represents a very low extra costin terms of processing. This cost reduction is possible because the algorithm thatperforms the initial division of the region of interest needs the data surface to workwith; taking advantage of this need, a copy of the file with the data representingthe variables of interest and their coordinates its made accessible to the web serverapplication, which utilizes it to plot the data surfaces using JavaScript. In thefuture, the web server will also allow the user to query the database and receivefiltered data through tables and charts.



6 Conclusions and future work

Pollution-Spots is a PS-based pollution monitoring system that proposes a cost-effective solution to overcome the absence of sensors to measure air qualityin mobile devices, with no extra cost related to data transport. By exploitingthe fixed sensor network and user participation, the system requires little usercontribution: the user only needs to have the application installed in her cellphone,thus eliminating the downside of carrying the sensing devices. Pollution-Spotsalso improves the spatial and time resolution when compared to the traditionalmonitoring systems (owned and managed by county agencies), tackling andsolving this important societal problem.

Pollution-Spots is still under development and further experiments andimprovement will be done based on the results of the first deployment in thecampus of the Pontificia Universidad Javeriana. The complete implementation ofthe scheme will provide us with more realistic information about its performance,scope and weak points. A critical issue is the trade off between privacy protectionand energy consumption that represents the use of Hybrid privacy preservingmechanisms. In order to determine the characteristics and specifications of thismechanism, it is necessary to model the most representative scenarios where theuser information could be compromised.

References

[1] Mendez, D., Perez, A., Labrador, M. & Marron, J., P-sense: A participatorysensing system for air pollution monitoring and control. Proc. of IEEEInternational Conference on Pervasive Computing and Communications,WiP workshop, pp. 344–347, 2011.

[2] Carrapetta, J., Youdale, N. & Chow, A., Haze watch, 2010.[3] Dutta, P., Aoki, P.M., Kumar, N., Mainwaring, A., Myers, C., Willett, W. &

Woodruff, A., Demo abstract: Common sense: Participatory urban sensingusing a network of handheld air quality monitors. Proceedings of the 7thACM Conference on Embedded Networked Sensor Systems, 2009.

[4] Hedgecock, W., Volgyesi, P., Ledeczi, A. & Koutsoukos, X., Disseminationand presentation of high resolution air pollution data from mobile sensornodes. Proceedings of the 48th Annual Southeast Regional Conference, ACMSE’10, pp. 6:1–6:6, 2010.

[5] Devarakonda, S., Sevusu, P., Liu, H., Liu, R., Iftode, L. & Nath, B., Real-time air quality monitoring through mobile sensing in metropolitan areas.Proceedings of the 2Nd ACM SIGKDD International Workshop on UrbanComputing, UrbComp ’13, pp. 15:1–15:8, 2013.

[6] Li, N., Jiang, Z., Liu, Z. & Meng, X., A method of hierarchical time-seriesdata visualization. Proceedings of the 6th International Symposium on VisualInformation Communication and Interaction, VINCI’13, pp. 113–114, 2013.

[7] Ourcast, Minutely.the minute-by-minute rain forecast at your exact location,2013.



[8] Vergara-Laurens, I., Mendez, D. & Labrador, M., Privacy, quality ofinformation, and energy consumption in participatory sensing systems.Pervasive Computing and Communications (PerCom), pp. 199–207, 2014.

[9] Vergara-Laurens, I., Mendez-Chaves, D. & Labrador, M., On the interactionsbetween privacy-preserving, incentive, and inference mechanisms inparticipatory sensing systems. Network and System Security, Springer BerlinHeidelberg, volume 7873 of Lecture Notes in Computer Science, pp. 614–620, 2013.

[10] Lands Nathan, B.J., Gamification wiki, 2010.[11] Han, K., Graham, E., Vassallo, D. & Estrin, D., Enhancing motivation in a

mobile participatory sensing project through gaming. Privacy, Security, Riskand Trust (PASSAT), pp. 1443–1448, 2011.

[12] Ueyama, Y., Tamai, M., Arakawa, Y. & Yasumoto, K., Gamification-basedincentive mechanism for participatory sensing. Pervasive Computing andCommunications Workshops (PERCOM Workshops), pp. 98–103, 2014.

[13] Zhang, J., Ma, J., Wang, W. & Liu, Y., A novel privacy protection schemefor participatory sensing with incentives. Cloud Computing and IntelligentSystems (CCIS), volume 03, pp. 1017–1021, 2012.



pollutionspots:anovelmethodfor air pollution monitoring · bogota, colombia abstract ... means of...

Documents